IDyOT: A Computational Theory of Creativity as Everyday Reasoning from Learned Information - Computational Creativity Research: Towards Creative Machines - page 130

(a)

Generator
1

Selection

.
.
.
.
.
.

Global Workspace/
Conscious Awareness

Associative Memory
with adaptive
representations

Generator
n

(b)

Generator
1

Buffer

to selection

match

Distribution

from perceptual input from memory

Fig. 7.1 Overview of the IDyOT architecture. a Generators synchronised to perceptual input sam-
ple, given previously buffered perceptual input (if any), from a mixed-order, multidimensional
Markov model to predict the next symbol in sequence, which is matched with the input. Predicted
symbols that match are buffered by each generator until it is selected by the throttling function. The
selected generator then flushes its buffer into the Global Workspace (“conscious awareness”). This
allows the resulting chunk of sequence to be stored in the memory, to become part of the statistical
model and thence to be used subsequently. b Within the generators, perceived inputs are matched
probabilistically with predicted distributions, and then the match that best matches expectation,
from the generator's prediction, is entered into a buffer. The buffer flushes when the generators'
buffer content is selected by the throttling function

where a match, or a prediction from a match, contains a large amount of information
in comparison with other current matches, the corresponding generator's buffer is
emptied into the Global Workspace, displacing what was there previously (though
what was there is stored in the memory, as below). Buffers that fail to empty are
forgotten after a fixed period, the duration of that period, alongside other potential
reasons, being a question of the research. The model entails that, for perception to
work, at least some generators must be working in all perceptual modalities at all
times, including sensory ones; otherwise nothing would be predicting for new input